Environment for Development
Discussion Paper Series
December 2014

EfD DP 14-22
The Land Certification
Program and Off-Farm
Employment in Ethiopia
Mi nt ew ab B ez abi h , An d re a Ma nn berg , an d E ye ru sal em Si b a
Environment for Development Centers
Central America
Research Program in Economics and Environment for Development
in Central America
Tropical Agricultural Research and Higher Education Center
(CATIE)
Email: centralamerica@efdinitiative.org
Chile
Research Nucleus on Environmental and Natural Resource
Economics (NENRE)
Universidad de Concepción
Email: chile@efdinitiative.org
China
Environmental Economics Program in China (EEPC)
Peking University
Email: china@efdinitiative.org
Ethiopia
Environmental Economics Policy Forum for Ethiopia (EEPFE)
Ethiopian Development Research Institute (EDRI/AAU)
Email: ethiopia@efdinitiative.org
Kenya
Environment for Development Kenya
University of Nairobi with
Kenya Institute for Public Policy Research and Analysis (KIPPRA)
Email: kenya@efdinitiative.org
South Africa
Environmental Economics Policy Research Unit (EPRU)
University of Cape Town
Email: southafrica@efdinitiative.org
Sweden
Environmental Economics Unit
University of Gothenburg
Email: info@efdinitiative.org
Tanzania
Environment for Development Tanzania
University of Dar es Salaam
Email: tanzania@efdinitiative.org
USA (Washington, DC)
Resources for the Future (RFF)
Email: usa@efdintiative.org
The Environment for Development (EfD) initiative is an environmental economics program focused on international
research collaboration, policy advice, and academic training. Financial support is provided by the Swedish
International Development Cooperation Agency (Sida). Learn more at www.efdinitiative.org or contact
info@efdinitiative.org.
Environment for Development
Bezabih et al.
The Land Certification Program and Off-Farm
Employment in Ethiopia
Mintewab Bezabih, Andrea Mannberg, and Eyerusalem Siba
Abstract
Land tenure security has long been touted as key to increased performance of the agricultural
sector in developing countries. At the same time, off-farm employment is seen as a strategy to diversify
rural economies. This paper utilizes household level panel data to analyse the impact of a land
certification program on farmers’ off-farm participation and activity choices in the Central Highlands of
Ethiopia. Identification of the program’s impact relies on the sequential nature of its implementation and
application of the Difference-in-Differences strategy. Our results suggest that certification is a
significant determinant of participation in off-farm employment. However, the impact differs
substantially between different types of off-farm activities. While land certification is associated with an
increased probability of participation in non-agricultural activities requiring unskilled labor, it reduces
the probability of engaging in work on others’ farms. In addition, the effect of the program depends on
the size of landholdings. The differences in the responsiveness of different off-farm activities to both
certification and farm size indicate the need to recognize the complex relationships between reform
policies that enhance land tenure and the non agricultural sub-sector in rural areas. In light of similar
previous studies, the major contributions of the paper are twofold: assessment of the effects of enhanced
land tenure security on activities outside agriculture and evaluation of the role of farm size in
determining off-farm participation.
Key Words: off-farm employment, land certification, farm size, Ethiopia
JL Codes: Q15, Q18, C35
Discussion papers are research materials circulated by their authors for purposes of information and discussion. They have
not necessarily undergone formal peer review.
Contents
1. Introduction ......................................................................................................................... 1
2. The Land Tenure System in Ethiopia and the Land Certification Program:
Background ....................................................................................................................... 4
2.1 Evolution of Land Tenure Policies in Ethiopia............................................................. 4
2.2 The Land Certification Program in Ethiopia ................................................................ 5
3. Data and Descriptive Statistics .......................................................................................... 7
3.1 The Certification Variable ............................................................................................ 8
3.2 Participation in Off-Farm Activities and Off-farm Employment Activity Choice ....... 8
3.3 Other Control Variables ................................................................................................ 9
3.4 Tenure Security Variables........................................................................................... 10
4. Empirical Strategy ............................................................................................................ 10
4.1. The Propensity Score Matching Method ................................................................... 11
4.2 The Difference-in-Differences Method ...................................................................... 12
4.3 Randomness of Treatment and Control Kebeles and the Common Trend Assumption14
5. Results ................................................................................................................................ 15
5.1 The Propensity Score Matching Results ..................................................................... 15
5.2 Off-farm Employment Participation Results .............................................................. 16
5.3 Off-farm Employment Activity Choice Results ......................................................... 18
5.4 Robustness Checks...................................................................................................... 19
Testing for the Common Trends Assumption ..............................................................19
6. Conclusions ........................................................................................................................ 20
References .............................................................................................................................. 23
Figures and Tables ................................................................................................................ 26
Appendix Figures .................................................................................................................. 38
Appendix Tables.................................................................................................................... 40
Environment for Development
Bezabih et al.
The Land Certification Program and Off-Farm
Employment in Ethiopia
Mintewab Bezabih, Andrea Mannberg, and Eyerusalem Siba1
1. Introduction
Small-scale agriculture is a major employer in many poor countries. Ethiopia is no
exception. The heavy dependence on small holder and low productivity agriculture makes a
large share of the population vulnerable to weather and production-related shocks. Land
certification has been one strategy to increase productivity. In addition, diversification of
livelihood, particularly toward off-farm activities, has been suggested as a means of reducing
vulnerability, as well as transforming the overall economy (Ellis 2000; Barrett et al. 2001;
Lanjouw and Lanjouw 2001). This paper evaluates the extent to which a land certification
program in Ethiopia impacts participation in off-farm employment and activity choice of
rural households in the central highlands of Ethiopia.
Participation in off-farm activities may be driven by both push and pull factors.
Households may be pushed into alternative livelihoods due to food insecurity, for example, or
pulled into such activities due to demand from other sectors.2 A household’s ability and need
to engage in off-farm activities thus depends to a large extent on its endowment of labor, land
and other productive resources (Woldehana and Oskam 2001; Holden et al. 2004; Shi et al.
2007). Farmers’ ability to participate in off-farm employment may also be determined by
1
Bezabih (corresponding author: mintewab.bezabih@port.ac.uk), London School of Economcs. Mannberg:
Umeå University, Sweden. Siba: University of Gothenburg, Sweden. Financial support from the Global Green
Growth Institute, the Grantham Foundation for the Protection of the Environment and the Economic and Social
Research Council (ESRC) through the Centre for Climate Change Economics and Policy, the Söderbergska
Foundation and the Wallander-Hedelius Fund is gratefully acknowledged. Access to data used for the analysis
was kindly granted by the Environmental Economics Policy Forum for Ethiopia, under the Environment for
Development Initiative. This work has also benefited from comments of Henrik Hansen, John Rand, Carol
Newman, Edward Samuel Jones, participants at the Annual World Bank Conference on Land and Poverty, 2013,
and seminar participants of the Development Economics Research Group (DERG), University of Copenhagen.
The usual disclaimers apply.
In settings where farming is associated with low profits and high risk, “distress-push” diversification may
motivate off-farm participation. In such instances, the non-farm sector serves as insurance against diminishing
returns to assets and as risk management in the agricultural sector (Barrett et al. 2001; Block and Webb 2001;
Rijkers and Söderbom 2013). Diversification due to “demand-pull” factors, on the other hand, is often motivated
by seasonal and interpersonal aggregation of household income and consumption, as well as economies of scope
in livelihood diversification (Davis 2003). In addition, participation in off-farm employment may depend on the
ability to participate, e.g., depending on human and physical capital endowments (Deininger and Olinto 2001;
Escobal 2001; Woldehana and Oskam 2001; Van den Berg and Kumbi 2006; Lanjouw et al. 2007; Bezu and
Barrett 2012).
2
1
Environment for Development
Bezabih et al.
tenure security, because leaving the farm may be associated with a risk of losing the land in
settings where the land system is characterized by tenure insecurity (Do and Iyer 2008;
Deininger et al. 2008; Jin and Deininger 2009).
Ethiopia’s current land tenure system makes for an ideal case to study the effect of
improved tenure security on off-farm participation, for two main reasons. First and foremost,
the land rights for Ethiopia’s farming masses have been associated with inherent tenure
insecurity, partly caused by the fact that farmers hold only usufruct rights to land and all land
is formally owned by the state (Crewett et al. 2008). Second, the country has recently
implemented a land certification program aimed at reducing tenure insecurity resulting from
usufruct land rights.
Accordingly, the central hypothesis of this paper is that land certification enhances
tenure security, which in turn enhances participation in off-farm employment. Tenure
security may affect incentives to engage in off-farm activities, both directly and indirectly.
The direct effect consists mainly of a reduction in the expected cost of being away from the
land (i.e., a reduction in the risk of land loss through redistribution).3 The indirect effects
operate via farm level intensification (e.g., investments in soil conservation (Holden et al.
2009; Deininger et al. 2011) and increased use of external inputs (Holden and Yohannes
2002). While the direct effect is expected to be positive, farm level intensification may affect
off-farm participation both positively and negatively. On the one hand, farming
intensification may lead to an increased need for labor on the farm, and this naturally reduces
the amount of labor available to engage in off-farm activities. If investment in the farm
increases efficiency, on the other hand, this may free up labor available for participation in
off-farm employment. In addition, we hypothesize that the impact of certification varies
depending on the type of off-farm activities. We base this hypothesis on the fact that the
different off-farm activities have differing relationships with on-farm activities (affecting onfarm labor demand), as well as varying needs for physical proximity to the farmer’s own farm
(affecting concerns about land redistribution, discussed above).
In the empirical study, we use survey data obtained from the Amhara National
Regional State of Ethiopia, containing information on household level off-farm employment
participation, to estimate the effect of the land certification program on off-farm participation
and on individual off-farm activities. We further analyze the effects of farm size and
3
If off-farm activities are interpreted as a signal of excess landholdings by the government, and if the
government has the right to redistribute land, farmers may avoid productive off-farm activities. If land
certification reduces the insecurity related to being away from the farm, we thus may expect an increase in
productive off-farm engagements, as individuals would no longer be constrained by the repercussions of tenure
insecurity.
2
Environment for Development
Bezabih et al.
availability of adult labor. The empirical strategy follows the Difference-in-Differences
method widely employed in impact assessment studies and exploits the gradual
implementation of the certification program as an identification strategy.
Our study is closely related to Zhang et al. (2004) and to Jin and Deininger (2009),
who analyze the relationship between tenure security and off-farm employment in China. It is
also related to Deininger et al. (2008) and to Do and Iyer (2008), who conduct a similar
analysis on Indian and Vietnamese data, respectively. Jin and Deininger (2009) use
household panel data to analyze the link between access to land rental markets and
occupational diversification. Their results suggest that land rental markets have a positive
effect on engagement in off-farm activities. Similar results are found in Zhang et al. (2004)
and in Deininger et al. (2008).
In contrast to the above-mentioned studies, our study is primarily interested in offfarm employment as an outcome variable. We also employ a measure of tenure security that
is more general than increased activity in rental markets. The study most closely related to
ours is by Do and Iyers (2008), who investigate the impact of a land certification program on
the time spent on non-farm activities. Like our study, their identification strategy relies on the
non-uniform timing of the land certification across the country.
Our study is closely related to a few previous studies investigating the association
between tenure security and non-farm employment (Jin and Deininger (2009), Zhang et al.
(2004), and Deininger et al. (2012) on China; Deininger et al. (2008) on India; and Do and
Iyers (2008) on Vietnam). Using a household panel dataset, Jin and Deininger (2009)
investigate the contribution of a land rental market to occupational diversification, where the
likelihood of renting out land is associated with income diversification and migration in
China. The positive relationship between off-farm employment opportunities and the
likelihood of renting out land is also found in Zhang et al. (2004). Exploiting cross-state
variation in land rental restrictions, using a household panel dataset, Deininger et al. (2008)
find that the land rental market is more active in locations with high levels of non-farm
activities in India.4 Unlike the above studies, our study is primarily interested in non-farm
employment as an outcome variable and employs a measure of tenure security that is more
general than increased rental markets, which is only one of the many outcomes of land
certification programs. Moreover, our study looks into the impact of certification on activity
choice, an issue not covered in previous studies. Our analysis of certification at both a village
4
Lohmar (1999) also found that tenure insecurity, as measured by percentage of households with land changes
due to village-wide relocation, is negatively associated with probability of participation in off-farm activities.
3
Environment for Development
Bezabih et al.
and household level is also uncommon in previous studies, which look into certification at a
province level only.
The study most closely related to ours is by Do and Iyers (2008), who investigate the
impact of a land certification program on the time spent on non-farm activities using the DiD
approach. Like our study, their identification strategy relies on the non-uniform timing of
land certification across the country. The role of tenure security for off-farm employment
participation is largely overlooked in the African literature. We also refine the study of
participation in the non-farm sector by looking at various activity choices within the sector.
The rest of the paper is organized as follows. Section 2 provides background
information on the development of land tenure policies and the certification program in
Ethiopia. Section 3 provides information on the datasets used for the analysis and discusses
descriptive evidence. Section 4 provides methodological discussion on the estimation strategy
and presents the econometric models used. Baseline results together with their robustness
checks are presented in Section 5. Finally, Section 6 concludes and discusses policy
implications of the study.
2. The Land Tenure System in Ethiopia and the Land Certification Program:
Background
2.1 Evolution of Land Tenure Policies in Ethiopia
Under the feudalistic system characterizing Ethiopia’s political and economic
landscape until 1975, the elite held all land and farmers’ tenure security hinged upon the
quality of tenant-landlord relations. The feudalistic system was ended and replaced by a
socialist state (Derg) in 1975 with the overthrow of Emperor Haile Selassie. Under the Derg,
all land was nationalized and redistributed to farmers via peasant associations.5 Land rights
were defined on a usufruct basis with no transfer rights to sell, lease, mortgage or sharecrop.6
Due to increasing population pressure, government land redistribution was frequently
implemented, and the maximum landholding per family was set at ten hectares. Eligibility
and access to land was contingent on a physical presence on the land. Taken together, the
system was characterized by a very high level of tenure insecurity, including a ban on land
rental activities that is believed to have effectively prevented any migration or pursuit of
alternative livelihoods for rural landholders (Kebede 2002; Adnew and Abdi 2005).
5
Peasant associations are local level administrative organs mandated to handle land related matters
6
Transfer via inheritance was allowed, but only to immediate family members and only if permission from the
peasant association had been acquired.
4
Environment for Development
Bezabih et al.
In 1991, the Derg was overthrown and replaced by the Ethiopian People’s
Revolutionary Democratic Front (EPRDF). The 1995 constitution introduced a few
improvements in tenure security, e.g., a requirement that farmers exposed to expropriation
have the right to compensation for any investments made on the land and an extension in the
time farmers are allowed to lease out land. However, the new government, to a large extent,
maintained the land policy of the Derg: land was defined as public property, which farmers
were forbidden to sell or to use for other means of exchange, and farmers were required to
leave the land if it was needed for public purposes. As a consequence, the 1995 constitution is
believed to have only partially resolved the tenure insecurity problems of rural land holders
(Adnew and Abdi, 2005).
In an effort to enhance tenure security and improve utilization of land, further steps
were taken by the federal and regional governments of Ethiopia. In particular, the Rural Land
Administration and Use Proclamation was drafted in 1997 (and revised in 2005 by the
Ministry of Agriculture and Rural Development) to give farmers who hold title certificates
the right to pass land rights on to their family members, to lease out plots to other farmers and
investors based on the land administration rules without being displaced, and to use land as
collateral (Adnew and Abdi, 2005). In 2002, the regional states received greater legislative
powers enabling them to form their own land-related policies, and new regional structures for
land administration (such as the Environmental Protection Land Use and Administration
Authority, EPLUA) were established. The main objectives of the certification program were
to improve tenure security through land registration and titling, to promote better land
management and investment, and to reduce conflicts among farmers over land boundaries
and user rights.
2.2 The Land Certification Program in Ethiopia
The Ethiopian land certification program was first initiated in the Tigray region in
1998, followed by the Amhara region in 2003/2004. Oromia and SNNP started certification
in 2007. The program is a variant of the land legislation programs that many African
countries have been implementing since the 1990s to remedy some of the perceived
shortcomings of existing systems. It differs from traditional land reform programs in terms of
the relatively low cost at which it has been implemented and the participatory nature of the
program. Between 1998 and 2007, it has been estimated that over 5 million farm households
have certified their land holdings in the four regions of Ethiopia (Deininger et al. 2011;
Adnew and Abdi 2005). The cost has been estimated to be about 1 USD per farm plot or 3.5
USD per household (Deininger et al. 2008). Given that conventional titling costs up to about
150 USD per household in Madagascar (Jacoby and Minten 2007), the Ethiopian program
can indeed be argued to be low-cost. The certificates ascribe farmers with written user rights
to demarcated pieces of land. Before the final certificate is issued, the document is subject to
5
Environment for Development
Bezabih et al.
a series of negotiations between participating farmers and implementing officials (Deininger
et al. 2011).
The implementation of the program is conducted through Land Administration
Committees (LACs) at the woreda7 level in five distinct steps. First, an awareness raising
meeting regarding the purpose and organization of land registration and certification is
conducted between the woreda and kebele administration and farmers (Palm 2010). In the
second step, Land Administration and Use Committees (LACs) are elected, and the elected
LAC members are trained. During the third step, individual households’ plots are identified
and demarcated jointly by LAC members, the designated household and households with
neighboring fields (Adnew and Abdi 2005). In the fourth step, the registered information is
entered on the forms and any outstanding conflict is passed to the courts; the result of the
land adjudication is then presented to the public for a month-long verification in order to
allow for corrections. In the final step, the book of holdings is registered. The legal status of
the holding is registered by the woreda EPLAUA head together with the LAC chairperson
(Olsson and Magnérus 2007; Palm 2010). Certified households are provided with a document
which typically includes the names and photographs of the household head and spouse, the
size and location of the land holding, and the neighbors of the demarcated land of the
household.
Previous impact evaluation studies of the certification program indicate that the
program has boosted farmers’ perceptions of tenure security and has improved land rental
market participation of female-headed farm households (Deininger et al. 2011; Holden et al.
2011; Bezabih et al. 2011). The program’s impact has been assessed in relation to agricultural
productivity (Holden et al. 2009; Deininger et al. 2011; Bezabih et al. 2012) and land related
investments such as soil and water conservation (Holden et al. 2009; Deininger et al. 2011).
However, to our knowledge, no study has previously analyzed the extent to which the land
certification program has affected rural households’ participation in off-farm activities.
7
The kebele is the smallest administrative unit in Ethiopia, while the woreda is the next largest, formed of a
collection of kebeles.
6
Environment for Development
Bezabih et al.
3. Data and Descriptive Statistics
To implement the analysis, we use panel data collected through a set of rural
household surveys, collected during the 2005 and 2007 cropping seasons.8 The surveyed
households are located in two zones (South Wollo and East Gojjam) of the Amhara National
Regional State, a region that encompasses part of the northern and central highlands of
Ethiopia. The choice of the two zones is intended to reflect the agro-ecological diversity
within the region, with East Gojjam having high agricultural potential and less rugged
topography, and enjoying a more reliable rainfall pattern, than South Wollo. The sample
consists of a panel of 1720 households (about 120 from each Kebele) randomly selected in 14
Kebeles (seven from each zone)910.
The rural household survey was designed independently of the implementation of the
certification program, and therefore includes households that were covered by the
certification program and those that were not. In addition, the first three rounds of the survey
were effectively conducted before the implementation of the program. This makes these
rounds an ideal baseline for estimating the effects of the program. The last round of the
survey, conducted in 2007, was designed to gather information on the features of the
certification program that enable analysis of the impact of the program on different variables
of interest. As noted in Section 2, the certification program was introduced in the region in
2003/2004, although full scale implementation started in 2005. The survey contains detailed
data on off-farm employment participation, agricultural production, physical farm
characteristics, indicators of tenure security and details on the process of land certification.
The construction of the certification, off-farm employment and other explanatory variables
are discussed below.
8
This multi-year survey was conducted by the Ethiopian Development Research Institute and Addis Ababa
University, in collaboration with the University of Gothenburg, and with financial support from the Swedish
International Development Cooperation Agency (Sida) since 2000. To date, four rounds of the Ethiopian
Environmental Household Survey (EEHS) have been collected, during 2000, 2002, 2005, and 2007. The last
two rounds of the survey will be used in the main analysis, while part of the information in the first two rounds
is used for robustness checks, as discussed in the subsequent sections.
9
Due to the panel nature of the data, attrition bias could be a potential problem. However, a simple inspection of
the data showed that attrition may not be a problem, as the households covered in the last two survey rounds
have been fairly similar.
10 247 out of the 2,617 households (9.44 percent) surveyed in 2005 did not participate in the last round of the
survey. Of these, 82 households were living in kebeles reached by the certification program and 165 were living
in kebeles not reached by the program before the last round of the survey. Our tests for attrition bias show that
there is a significant difference in the risk of attriting in kebeles reached and not reached by the certification
program. However, we find no evidence of attrition bias in terms of our outcome variable. To test for attrition
bias, we estimate a Heckman selection model where the first step is selection into staying in the sample and the
second step estimates the probability of engaging in off-farm activities corrected for selection bias. The
coefficient on the inverse mills ratio was insignificant in all specifications.
7
Environment for Development
Bezabih et al.
3.1 The Certification Variable
The certification program is characterized by a gradual rollout, where certification in
each kebele occurs in a rapid, campaign-style. The gradual rollout implementation was
adopted because the initial plan of simultaneously reaching all woredas (and kebeles) within
the region could not be realized.11 Our dataset contains information on whether or not a
household had a certificate of its landholdings 12 months prior to the last survey round. This
makes it possible to define the intervention at a household level. However, many of the
households that are marked as “non-certified” in the data are in reality waiting for their
formal documents and have been excluded for temporary reasons such as shortage of papers
or delays in their program registration. Indeed, in villages where a considerable number of the
households were not certified at the time of the last survey or were not certain that they
would be certified, all were eventually certified. Because spillover effects are likely, defining
certification at a kebele level may also be justified.
We therefore use two means of identifying certified households: at the household and
kebele levels. If a given household had acquired certification at least 12 months before the
last round of the survey (in 2007), the household is labeled as “certified” according to a
household level definition of certification. For the kebele level certification, we define a
dummy certification variable, which takes the value of one if the kebele to which a given
household belongs was reached by the program 12 months prior to the 2007 survey round and
a value of zero otherwise. It should be noted that, because most of the control group had
already received information about the program or undergone the registration process, our
estimates reflect a lower bound of the total effect of the program (Deininger et al. 2011).
3.2 Participation in Off-Farm Activities and Off-farm Employment Activity
Choice
Our off-farm employment variables distinguish between participation and activity
choice measures. Participation in off-farm activities is a binary variable, which represents the
outcome of participation in any off-farm employment activity. The variable takes the value of
one if the household engages in any off-farm activity, and the value zero if there was no
participation in any such activity. The off-farm activity choice variables are represented by 14
different activities. However, due to low frequencies in some activities, we have merged the
categories into five. The first two categories include farm worker (agricultural activities on
other people’s farms) and free worker (labor sharing arrangements where people contribute
labor freely but also expect the favor back in terms of labor contribution). The third category,
11
This is reported to be due to shortages in financial and manpower resources both at kebele and woreda levels.
8
Environment for Development
Bezabih et al.
professional work, includes teaching, mechanics, driving, clerical jobs, administrative work,
health work, building and crafts making. In the fourth category, unskilled off-farm activity
includes household help, shop keeping, security guard, and other miscellaneous activities.
The last category is food-for-work – a program where participants are involved in public
infrastructural development such as building conservation structures, dry-weather roads, or
tree planting, and are paid in kind. Increases in farm work, free work, unskilled work and
professional work due to the certification program are considered positive outcomes, because
these activities are labor diversification strategies likely to improve household welfare.
However, increased participation in food-for-work activities is less likely to be an intended
effect of the program, as engagement in this type of activity points instead to a vulnerability
of the household.
Table 1 presents the distribution of individuals across different off-farm activities in
2005 (before the program) and in 2007 (after the program). The first two columns are for the
pooled sample, the second two are for kebeles reached by the program in 2007 and the last
two are for kebeles not reached by the program in 2007. As Table 1 shows, most households
do not engage in off-farm activities, but there seems to be a trend toward more engagement.
Around 88 percent of the households did not engage in any off-farm activities in 2005. In
2007, this figure had fallen to 79 percent. Table 1 further reveals that the trends across
different activities are very different. While engagement in professional work increased
substantially during the study period (from 0.82 percent to 4.79 percent), participation in farm
work outside of one’s own farm decreased (from 5.16 percent to 1.9 percent). It is also
notable that engagement in food-for-work activities increased from 0.14 percent in 2005 to
9.57 percent in 2007.
Although relatively more individuals in certification villages engaged in off-farm
activities (15.7 percent) than in non-certification villages (9.3 percent), the structure of
participation in the different off-farm activities was relatively similar in certification and noncertification villages in 2005 (i.e., prior to certification). In 2007, the difference in off-farm
participation (all activities) is smaller, but, if we look at the individual activities, we see that
the pattern is slightly different in certification and non-certification villages. Participation in
farm work on others’ land and free work fell much more in certification villages than in noncertification villages, while participation in professional work, unskilled work and food-forwork increased more.
3.3 Other Control Variables
The relevant summary statistics for our control variables are presented in Table 2. The
first panel of the table contains summary statistics for the pooled sample, while the second
and third panel depict statistics for certified and non-certified villages, respectively. The
statistics presented are for the entire time period (2005 and 2007).
9
Environment for Development
Bezabih et al.
As can be seen in Table 2, the average age of the household head is about 51 (49 in
certified villages and 51 in non-certified villages) in the sample. 14 percent of the surveyed
households have a female head of household. The share of female-headed households is
slightly higher in treated villages (16 percent) compared to non-treated villages (12 percent).
About half of the households have an illiterate household head (51 percent and 56 percent in
certified and non-certified villages, respectively). Households have, on average, two females
and two males of working age (above 15 years), 4.6 livestock (tropical livestock units) and
1.8 oxen. Households in certified villages hold slightly more livestock and oxen than
households in non-certified villages.
Because physical farm characteristics may influence labor demand on the farm, we
also include variables related to soil and land. As can be seen in Table 2, households in our
sample have user rights to 2.5 hectares of land on average. The mean plot area is 0.3 hectare.
About 56 percent of the plots in the sample are defined as fertile, while 10 percent of the
plots are located on infertile soil. The share of plots with infertile soil is slightly higher in
kebeles not reached by the certification program (11 percent versus 8 percent in certification
kebeles). Finally, a majority of the plots are located on flat land (71 percent). However, a
slightly higher share of plots in non-certified kebeles is located on steep slopes (6 percent
versus 2 percent).
3.4 Tenure Security Variables
In order to measure tenure security, we use experience of land loss and expected
changes in land holdings in the coming five years.12 In kebeles reached by the certification
program in 2007, 6 percent had experienced land loss. This figure is higher in our control
group, where 10 percent had lost land in the 5 years preceding the surveys. 14 percent expect
to lose land in the future (13 percent in certification kebeles and 15 percent in the control
group), while 32 percent expect no change in landholdings (36 percent and 29 percent in
certification and non-certification kebeles, respectively).
4. Empirical Strategy
In order to estimate the effect of the certification program on engagement in off-farm
activities, we combine the non-parametric approach of propensity score matching with the
12
The question is formulated as “In the coming five years, do you expect increase, decrease or no change in
uncertainty in your land holdings?” Experience of land loss is constructed as a dummy variable assuming a
value of 1 is there is any past experience with respect to changes in the size of land holdings, and zero
otherwise.
10
Environment for Development
Bezabih et al.
Differences-in-Differences (DiD) method. The essence of the propensity score estimation is
to balance the observed distribution of covariates across households in the control and
treatment groups, thereby reducing bias in the measurement of the program impacts that are
associated with observables (Rosenbaum and Rubin 1985). The DiD method is suitable for
identifying the effects of a random program intervention in which information on the
variables of interest exists before and after the introduction of the program. Such a program
typically targets a certain group of individuals (treated group) while the remaining group of
individuals (control group) is not exposed to the program (Wooldridge 2002). In our case, the
approach measures the impact of the land certification program by comparing the change in
off-farm participation and off-farm activity choice of households in certified
kebeles/households (treatment group) with the corresponding change for households in noncertified kebeles/households (control group).
However, the DiD approach is only a valid approach if the assumption of a parallel
trend in certified and non-certified kebeles/households is fulfilled,13 which could be
potentially invalidated by the presence of unobservable time variant differences. Because the
presence of such unobservables is inherently non-testable, we employ methods of indirectly
testing for the presence of common trends, or validating the so-called common-trend
assumption.
4.1. The Propensity Score Matching Method
In order to find a group of treated households/kebeles similar to the control
households/kebeles in all the relevant treatment characteristics, we start with estimating a
standard logit certification model, as given in Equation (1).
𝑃𝑖𝑡 = 𝛼 + 𝐻𝑖𝑡 + 𝑢𝑖𝑡
(1)
where, for household i and year t, 𝑃𝑖𝑡 is a dummy variable representing participation in the
program or not; 𝐻𝑖𝑡 is a vector of variables used as determinants of the likelihood of
participation; and 𝑢𝑖𝑡 is the error term, following a logistic distribution. The essence of the
logit equation is that it forms a basis for estimating the propensity score from predicted
values of Equation (1), which enables the generation of a comparison group by picking the
13
Violation of the common trend assumption implies that the effect of certification may not be identified
because there are underlying factors that cause the variable of interest to respond differently in the control and
treatment villages.
11
Environment for Development
Bezabih et al.
“nearest neighbor” with similar characteristics for each participant (Jalan and Ravallion
2003)14.
The propensity score is given by:
𝑒(𝑥) = Pr(𝑤 = 1)|𝑋 = 𝑥 = 𝐸(𝑤|𝑋 = 𝑥)(2)
where w is the indicator of exposure to treatment, and x is a multidimensional vector of pretreatment characteristics. The choice of covariates to be included in the propensity score
estimation is based on the principle of maintaining a balance in using common variables and,
at the same time, meeting the common support criteria (Jalan and Ravallion 2003). After
matching, there should be no systematic differences in the distribution of covariates between
the two groups; as a result, the pseudo-R2 upon matching should be fairly low and the joint
significance of all covariates should be rejected.
4.2 The Difference-in-Differences Method
The Difference-in-Difference analysis of impacts of certification on off-farm
participation is carried out on the matched sample generated from the propensity score
matching analysis discussed in Section 4.1. The hypothesized relationship between
certification and participation in off-farm employment is represented by Equation (3).
Pr(𝑅𝑖𝑡 = 1) = 𝛼𝑜 + 𝛼1 𝑃𝑖 + 𝛼2 𝑇𝑡 + 𝛼3 𝑃𝑖 ∗ 𝑇𝑡 + 𝛼4 𝑋𝑖𝑡 + 𝛼𝑖 + 𝜀𝑖𝑡
(3)
where 𝑅𝑖𝑡 is a dummy variable identifying the respondent household i’s off-farm participation
status at time t. 𝑃𝑖 is a dummy variable identifying whether or not the respondent household
is located in a treated kebele. 𝑇𝑡 represents a time dummy equal to one for the post-program
period and zero otherwise. The coefficient of the interaction variable 𝑃𝑖 ∗ 𝑇𝑡 thus captures the
impact of certification. Finally, 𝑋𝑖𝑡 is a vector of other control variables, including
socioeconomic and physical farm characteristics, with potential effects on off-farm
participation; 𝛼𝑖 represents time-invariant household specific characteristics; and the error
term is denoted by 𝜀𝑖𝑡 .
The parameters of interest in Equation (3) are α1 , α2 and α3 . α1 represents preexisting differences in off-farm participation between the treated and control groups in 2005,
while 𝛼2 represents the change in off-farm participation in the control and treatment villages
14
The nearest-neighbor matching method, compared to other weighted matching methods, such as kernel
matching, allows for identifying the specific matched observations that would be employed in the subsequent
difference-in-difference analysis.
12
Environment for Development
Bezabih et al.
between 2005 and 2007. Our major parameter of interest is 𝛼3 , which captures by how much
the likelihood of participation changed in the treated villages, as compared to the control
group.
We expand Equation (3) to incorporate the potential impact of farm size. The role of
farm size is represented by a variable that measures a household’s total landholdings and its
interaction with the certification variable (post-certification variable), denoted by 𝐹𝑖𝑡 and
𝑃𝑖 𝑇𝑡 ∗ 𝐹𝑖𝑡 , respectively. In Equation (4), the direct effect of farm size on off-farm
participation is captured by 𝛼4 , whereas 𝛼5 ≠ 0 captures heterogeneous impact of the
certification among holders of various sized farms due to the association between tenure
security and farm size.
𝑅𝑖𝑡 = 𝛼𝑜 + 𝛼1 𝑃𝑖 + 𝛼2 𝑇𝑡 + 𝛼3 𝑃𝑖 ∗ 𝑇𝑡 + 𝛼4 𝐹𝑖𝑡 + 𝛼5 𝑃𝑖 𝑇𝑡 ∗ 𝐹𝑖𝑡 + 𝛼6 𝑋𝑖𝑡 + 𝛼𝑖 + 𝜀𝑖𝑡
(4)
The probability of engaging in any type of off-farm activity is analyzed using the
standard random effects logit model, while the activity choice model is analyzed using the
multinomial logit model. As in the case of dichotomous logit models, the multinomial logit is
based on random utility theory. In other words, individuals who report that a member of their
household has participated in off-farm employment are assumed to choose the alternative
with the highest utility. In our analysis, individuals choose between five alternatives:
agricultural work on other people’s fields (daily labor), food-for-work, skilled off-farm work
and other off-farm employment. Accordingly, the multinomial logit specification is given in
(5).
𝐴𝑖𝑡 = 𝛽𝑜 + 𝛽1 𝑃𝑖 + 𝛽2 𝑇𝑡 + 𝛽3 𝑃𝑖 ∗ 𝑇𝑡 + 𝛽4 𝐹𝑖𝑡 + 𝛽5 𝑃𝑖 𝑇𝑡 ∗ 𝐹𝑖𝑡 + 𝛽6 𝑋𝑖𝑡 + 𝛽𝑖 + 𝜗𝑖𝑡
(5)
where 𝐴𝑖𝑡 is a variable representing the respondent household i’s off-farm activity choice
status at time t, 𝛽 represents the estimable coefficients and 𝜗𝑖𝑡 is the error term.
Our estimation strategy for both off-farm participation and activity choice is
associated with a number of challenges. First, the presence of household specific effects that
are not accounted for by the observed covariates could lead to omitted variable bias. In order
to correct for this potential bias, we follow Mundlak’s (1978) approach of incorporating the
relationship between the time varying regressors 𝑘𝑖𝑡 and the household fixed effect (  i for
the participation regression and 𝛽𝑖 for the activity choice regression). This enables controlling
for unobserved heterogeneity by adding the means of time varying covariates, also known as
the pseudo fixed effects or the Mundlak-Chamberlain’s Random Effects Model. In particular,
 i can be approximated by a linear function:
13
Environment for Development
𝛼𝑖 = 𝑎𝑘̅𝑖 + 𝜂𝑖𝑡 , 𝜂𝑖𝑡 ~𝑖𝑖𝑑(0, 𝜎𝜂2 )
Bezabih et al.
(6)
Substituting expression (6) in Equation (4) gives the estimable Equation in (7):
𝑅𝑖𝑡 = 𝛼𝑜 + 𝛼1 𝑃𝑖 + 𝛼2 𝑇𝑡 + 𝛼3 𝑃𝑖 ∗ 𝑇𝑡 + 𝛼4 𝐹𝑖𝑡 + 𝛼5 𝑃𝑖 𝑇𝑡 ∗ 𝐹𝑖𝑡 + 𝛼6 𝑋𝑖𝑡 + 𝑎𝑧̅𝑖 + 𝜓𝑖𝑡
(7)
A similar transformation into Equation (5) gives the estimable equation in (8):
𝐴𝑖𝑡 = 𝛽𝑜 + 𝛽1 𝑃𝑖 + 𝛽2 𝑇𝑡 + 𝛽3 𝑃𝑖 ∗ 𝑇𝑡 + 𝛽4 𝐹𝑖𝑡 + 𝛽5 𝑃𝑖 𝑇𝑡 ∗ 𝐹𝑖𝑡 + 𝛽6 𝑋𝑖𝑡 + 𝛽𝑖 + 𝑎𝑧̅𝑖 + 𝜓𝑖𝑡
(8)
4.3 Randomness of Treatment and Control Kebeles and the Common Trend
Assumption
The second estimation challenge is the proper identification of the program impacts
using the DiD estimator, which relies on the randomized choice of treatment and control
kebeles. In particular, if the choice of kebeles certified early in the process is systematically
related to factors that may affect our outcome variable, the measurement of the treatment
impact may be biased. In order to evaluate the degree of randomness in implementation, we
examined the criteria behind the choices. Correspondence with the administrative officials
from the EPLAUA confirmed that, in many cases, the kebeles were simply picked randomly
based on woreda administrative capacity. Hence, if administrative capacity is not strongly
associated with factors that affect off-farm participation, then the geographical discontinuity
in program implementation offers a valid identification strategy. To establish whether the
sampling of treated and control kebeles in the survey was random, a simple test was
conducted in terms of the difference in the location of kebeles relative to the main
road/nearest town. This variable could serve as a measure of remoteness, representing access
to information, technology, and markets. Accordingly, the average distance of the nearby
town from the treatment and control kebeles is calculated to be 69.5 and 72.5 minutes,
respectively, as per our survey data.15 The average distance from a nearby main road is also
about 24 and 37 minutes for the treatment and control kebeles, respectively. These figures
give no indication that the treatment and control kebeles differ significantly in terms of access
to information, technology and markets.
15
It should be noted that proximity to roads is a mere indicator of a set of confounding factors that affect the
decision to participate in off-farm employment. As we discuss below, the test for the common trend assumption
is a more formal way of assessing the randomness in the choice of the treatment and control kebeles. It should
also be noted that, given modern telecommunication technology, e.g., internet, social media etc., one may
question the role of physical distance, especially in access to information. However, modern communication
coverage remains minimal and even those “traditional” types of modern communication such as telephone
access remain road-side based.
14
Environment for Development
Bezabih et al.
While the random choice of kebeles is a necessary condition for accurately identifying
the program effects, the so-called common (or parallel) trend assumption forms the sufficient
condition. The essence of the DiD approach is to attribute any change over time in the
difference in outcomes between the control and treatment villages to the certification
program, having controlled for a range of explanatory factors. The common trend assumption
asserts that, in the absence of the treatment, one would expect parallel movement between the
treatment and control groups, essentially attributing the effect of changes in the treatment
group to the program only. An ideal test for this parallel movement would be to assess
whether, in the absence of the program, the trends in the two groups would have been
identical. However, this is unobservable, implying that the common trend assumption is
fundamentally untestable after the introduction of the program. As an alternative, it is
possible to test whether there were common trends in the treatment and control villages in the
period before the introduction of the program. To this effect, we used two methods:
comparing the average outcomes after controlling for a range of explanatory variables, and
testing for ‘placebo’ effects.
In order to do so, Equation (3) is estimated with different intercept terms for each year
and for each treatment category, using data for periods prior to the commencement of the
certification program. The presence of parallel trends is supported by the absence of
differences in the intercept terms between control and treatment villages over all the periods,
and vice versa for the lack of common trends. The results for the common trend assumption
test are presented in Section 5.4.
5. Results
In this section, we present the results of our empirical analysis. The section is outlined
as follows: we start by discussing the results of our matching procedure and then proceed by
discussing the decision to participate in off-farm activities. Finally, we display our results for
individual off-farm activities.
5.1 The Propensity Score Matching Results
The matching procedure is carried out using a one-to-one nearest neighbor matching
with Caliper. As discussed in Section 4.1, our matching procedure is based on a set of
relevant variables, which are presumed to affect the probability of both selection into
certification and off-farm participation. The variables included are: size of landholdings and
mean plot area, color of soil, fertility of soil, slope of plot, tropical units of livestock, number
of oxen, literacy of household head, number of male adults in the household and to what
extent the household has experienced land-related conflict and land loss. The choice of
variables is based on the socioeconomic or physical farm characteristics that potentially affect
selection into the program. The result of the logit estimation in the PSM matching procedure
15
Environment for Development
Bezabih et al.
is depicted in Table 3 below. As can be seen in the table, the most influential variables for
selection into a certification kebele is average plot area. In other words, in the unmatched
sample, households with larger average plots were more likely to live in kebeles that were
reached by the certification program.
The result of the matching procedure, in terms of a reduction in bias in systematic
selection into the program, is depicted in Figure 1 and Figure 2 below.16 As can be seen in the
figures, matching substantially reduces the bias.
The matched sample consists of a total of 2478 observations distributed over the two
years. This is the sample used for the empirical analysis below.17
5.2 Off-farm Employment Participation Results
We begin by examining the effect of the certification program on general off-farm
participation using a panel logit regression technique. Table 4 presents the results from the
analysis of the off-farm participation equation. Because our descriptive statistics suggest that
there may be different trends in agricultural and non-agricultural off-farm activities, we
estimate two regressions: one for participation in any off-farm activity and one for
participation in non-agricultural off-farm activities. Accordingly, Table 4 displays the results
for off-farm activities excluding work on others’ land in panel 1 (Columns 1-4), and the
results for participation in any off-farm activity (including farm work) in the second panel
(Columns 5-8). We present the results for four different models. Model I only contains plot
characteristics and socioeconomic variables as covariates. In Model II, we add controls for
living in a certification village before and after treatment, and in Model III we also include
farm size interacted with living in a certification village before and after treatment. Finally,
Model IV contains controls for different measures of tenure security. Because we are running
a random effects logit, marginal effects are not valid (since they assume that the individual
effect is zero). Consequently, Table 4 displays coefficient estimates.
As can be seen in the table, the definition of off-farm work matters to a significant
extent. If we include all types of off-farm work, i.e., including work on other households’
land, our estimation results suggest that certification (post_treatment) is associated with a
reduced likelihood of participating in off-farm activities (Columns 5-8). However, if we
remove farm work from our off-farm variables, we find that the certification had a positive
16
The results of the logistic regression on which the propensity scores are based and the common support results
are not reported but are available from the authors upon request.
17
See the appendix for the household level identification.
16
Environment for Development
Bezabih et al.
and strongly significant effect on participation in off-farm activities. This could be because
we have a significant share of the off-farm activities in the farm sector.
Our results reveal significant links between the size of landholdings and off-farm
participation, particularly when non-agricultural off-farm activities are considered. The link,
however, is insignificant when off-farm activities in the farm sector are included as off-farm
categories. This could be because the off-farm activities in the farm sector also include
unpaid work or shared labor, as well other semi-paid arrangements. The negative coefficient
of certification conditional on farm size indicates that large farm size would lead to reduction
in participation in off-farm participation, possibly due to the certification-related
intensification we discussed in the introduction. Alternatively, owners of smaller farm size
would be more likely to participate in off-farm activities after certification because of
increased tenure security (outweighing the intensification effects because the farm sizes are
smaller).
Turning to the tenure insecurity variables, we find that households with previous
experiences of land loss are more prone to engage in off-farm activities regardless of the
measure of these activities. This result may appear surprising at first sight: because control
over land is often correlated with continuous use of land, we may expect that households with
experiences of land loss would be more prone to stay on their land. However, land loss is also
likely to imply a reduction in land holdings, and thus that households with experiences of
land loss may find it necessary to work elsewhere.
Concerning the rest of the covariates, we find that female-headed households are less
likely to engage in off-farm work that includes work on other households’ land, whereas
households with a larger number of male and female adults are more likely to participate in
off-farm activities that exclude agricultural activities on other people’s farms. For
participation in off-farm activities excluding farm work, only the number of male adults
matters.
The results generally hold when defining the certification variable at the household
level. However, as can be seen in Table A1 in the appendix, the sample size is much smaller
in this case. This is because many households living in kebeles reached by the program in
2007 had not yet received their certificates at the time of the survey. Our results for the
household level identification suggest that, for off-farm activities excluding farm work, the
slope and soil characteristics are significantly correlated with the participation decision. More
specifically, we find that households that own a large share of plots with black or red soil are
more likely to participate than households that own plots with other color soil, while
households whose plots are located on steep slopes are less likely to participate. Holding
other factors constant, the importance of plot characteristics with respect to off-farm
participation could be associated with the respective labor demands. Accordingly, steep slope
17
Environment for Development
Bezabih et al.
plots may require more conservation work, reducing the availability of spare labor hours for
off-farm activities.
5.3 Off-farm Employment Activity Choice Results
In order to assess the impact of the certification program on activity choice, we also
estimate a multinomial logit regression. We cluster standard errors at the household level to
account for possible correlations in error terms. Table 5 presents the predictive margins
emanating from the estimation. Columns 1-5 display the predictive probability of being in
each activity. The reference group is those with no participation in off-farm employment.
The certification program is associated with an increased likelihood of participating in
unskilled non-agricultural work and in the food-for-work program, but a reduced probability
of engaging in professional work. We find no significant effects on the likelihood of
participating in paid farm work or in labor sharing arrangements (free work). The largest
effect found is for unskilled work in the non-agricultural sector. However, we also find a
relatively large effect on the likelihood of participating in food-for-work activities.
Concerning farm size, we find that the size of landholdings in itself reduces the probability of
engaging in food-for-work activities. However, we also find that certification significantly
reduces the probability of engaging in unskilled work for households with large landholdings.
Concerning the rest of the covariates, the results in Table 5 suggest that femaleheaded households are significantly more likely to have household members engaged in
professional work, while households with an illiterate head are significantly less likely to
have professionally active members. Households with a relatively larger number of male and
female adults are significantly more likely to engage in unskilled off-farm work and food-forwork activities. Households with many female adults are, however, less likely to engage in
professional work. Finally, we also find that households with a larger number of oxen tend to
engage in unpaid labor sharing agreements.
For the household level certification, the relatively small sample size made it
impossible to estimate the full multinomial logit (due to very few observations in some of the
categories). We were therefore forced to reduce the number of covariates and to only estimate
the probability of engaging in labor sharing arrangements, unskilled work and food-for-work
compared to no off-farm participation.18 The results suggest that certification reduced the
probability of engaging in unpaid labor sharing arrangements and increased the probability of
18
As can be seen in Table A2, we have removed square of age of the household head, share of red and black
soil, mean plot area, share of medium fertile soil and share of medium sloped plots from the analysis. These
variables were chosen for removal based on the lack of significance of their coefficients in other regressions.
18
Environment for Development
Bezabih et al.
engaging in unskilled work. No statistically significant effect on food-for-work activities was
found. However, this last finding may also have been a consequence of the small sample size.
5.4 Robustness Checks
Figure 3a and Figure 3b illustrate the underlying trend in off-farm employment
participation in treatment and control villages prior to the introduction of the program. The
figures plot the mean probability of engaging in off-farm activities, having controlled for a
set of explanatory variables.19 Figure 3a depicts the pattern for participation in off-farm
activities excluding farm work and shows a negative trend in off-farm participation both in
the certification and non-certification kebeles before the certification program commenced.
The negative trend is relatively stronger in certification kebeles; however, between 2005 and
2007, the trend in the kebeles reached by the program is reversed from negative to positive.
While this graphical illustration does not provide conclusive evidence of a common trend, it
lends some support to the hypothesis that the trend was not divergent before the program. The
trend for all types of off-farm activities is much less clear. Figure 3b depicts the average
tendency to engage in all types of off-farm activities for the years 2000-2007. In contrast to
the results in Figure 3a, Figure 3b shows a U-shaped trend in both certification and noncertification kebeles. In accordance with the estimation results in Table 2, the positive trend
after 2005 appears to be stronger in kebeles not reached by the certification program.
Testing for the Common Trends Assumption20
To test for any statistical difference in the trend between certification and noncertification kebeles, we also perform a test of “placebo effects.” We do this by limiting the
sample to the time period before the program started (in 2007 for our sample) and then
adding fake treatment variables to the regression. In other words, we use information in the
survey years 2000, 2002 and 2005 to examine whether there appear to be placebo treatment
effects amongst households in the treated villages before the program was introduced. The
estimated placebo treatment effects should, of course, ideally be close to zero and statistically
insignificant. The results of the placebo effect analysis are presented in Table 6. The first
panel in the table contains the result for off-farm activities excluding work on other
households’ farms, while the second panel depicts the results for participation in all off-farm
19
The dependent variable here is participation in off-farm employment, while the explanatory variables are
socioeconomic and physical farm characteristics, tenure security and participation in the certification program.
20
The number of observations for each activity is too small to conduct placebo tests for activity choice. As a
result, we rely on the common trends test results from the participation analysis.
19
Environment for Development
Bezabih et al.
activities. The individual columns represent different placebo tests. In each regression, we
use a full set of control variables, kebele fixed effects and Chamberlain-Mundlak effects.
As can be seen in the table, we do find significant placebo effects regardless of
whether we include farm work in the off-farm activity variable. However, while the off-farm
variable containing farm work produces a significantly positive placebo effect in 2005, thus
suggesting that a positive trend in certification kebeles started before the introduction of the
program, the coefficient for the no-farm-work regression produces a significantly negative
coefficient. In accordance with Figure 3a, this suggests that certification kebeles were on a
significantly more negative trend concerning no-farm off-farm activities than were noncertification kebeles. To see how the introduction of the true program affects the placebo
treatments, we also run a regression with both a placebo treatment and the true treatment
(post-treatment). The results are presented in Columns 4 and 8 of Table 6. As can be seen in
the table, the magnitude of the placebo effect falls, but the significance level remains the
same. Interestingly enough, the positive and significant effect for the off-farm regression,
excluding agricultural activities on other people’s farms, is robust to the inclusion of the
placebo variable.
The test results for the household level identification of certification are presented in
Table A3. Overall, these results are supportive of our causal interpretation of the kebele level
identification results.21 For the regressions on all types of off-farm participation, we find
placebo effects similar to those for the kebele level identification. However, for the no-farm
regression, the statistically significant placebo effect in 2005 disappears, while we find a
significantly positive effect in 2002. This suggests that the results at the household level are
somewhat less reliable.
6. Conclusions
Given the pivotal importance of the land tenure system in rural economic dynamics,
the land legislation programs implemented in many African countries since the 1990s have
received due attention in the literature. Specifically, there have been considerable efforts to
analyze the effects of such programs on agricultural productivity, land market participation
and investment across Africa (e.g., Pinckney and Kimuyu 1994; Jacoby and Minten 2007; Ali
et al. 2011).
21
It should be noted that Deininger et al. (2011) and Bezabih et al. (2012), using the same data used in this
study for their analysis, find no evidence of a difference in trends for soil conservation investment and
agricultural productivity, their respective variables of interest.
20
Environment for Development
Bezabih et al.
The Ethiopian Land Certification Program has spurred a considerable amount of
research, in particular on the impact of the program on agricultural investment and
productivity (e.g., Deininger et al. 2011; Holden et al. 2011, 2012; Bezabih et al. 2012).
However, to our knowledge, this is the first paper that assesses the effect of the program on
off-farm participation. More specifically, we estimated the effect of the land certification
program on participation in off-farm employment and on off-farm activity choice using data
from the Central Highlands of Ethiopia. We base our hypothesis on the positive link that has
been shown between the land certification program and tenure security (Bezabih et al. 2012).
Given that tenure insecurity induces aversion toward leaving the land in pursuit of
employment activities outside the farm, restoring tenure security could, in many situations, be
expected to have positive off-farm employment consequences.
The empirical results emanating from the panel logit off-farm participation and
multinomial logit activity choice models suggest that the effect of the certification program
on off-farm activities depends on the type of activity. We find that, while land certification
seems to have had a positive impact on the tendency to engage in unskilled non-agricultural
work, it may have had a negative effect on farm work on others’ land and on professional
work. We can only speculate on the mechanisms behind these results, but if the group of
households that consider engaging in unskilled off-farm activities are also the ones most
affected by tenure insecurity, then our results may point to a reduction in tenure insecurity
due to certification for these households. Our finding that the certification program had a
positive and significant effect on food-for-work activities is slightly disturbing, since foodfor-work cannot be said to be a productivity enhancing activity. However, if food insecure
households previously abstained from leaving their land due to fear of redistribution or land
conflict, and if the certification program contributed to these households engaging in
activities that provide them with food, this effect is not inherently negative.
The negative effect of the program on professional labor activities may be explained
by the fact that skilled and permanent off-farm employment activities possess characteristics
that make them relatively inelastic to exogenous policy changes in the short run. For instance,
skilled labor activities require investment in skills and working capital. Hence, the effect of
changes in incentives may show only in the longer run. Another possibility is rigidity in the
demand side of employment opportunities. While food for work activities have been able to
absorb labor in off agricultural seasons, there is some reason to believe that a more flexible
food-for-work employment program could increase off-farm employment rates during the
agricultural seasons. Potential rigidities associated with such seasonality, as well as limited
availability of skilled and non-farm employment opportunities, will make such activities
generally unresponsive to policy changes.
21
Environment for Development
Bezabih et al.
Taken together, our results suggest that the Ethiopian Land Certification Program has
affected engagement in off-farm activities. However, because the effect of the program seems
to vary across different types of off-farm activities, and because such effects may take time to
develop, there is a need for further studies on the subject as more data becomes available.
22
Environment for Development
Bezabih et al.
References
Ali, D. A., Deininger, K., and Goldstein, M. 2011. Environmental and Gender Impacts of
Land Tenure Regularization in Africa: Pilot Evidence from Rwanda. World Bank
Policy Research Working Paper Series.
Adnew, B., and F. Abdi. 2005. Land Registration in the Amhara Region, Ethiopia. Research
Securing Land Rights in Africa, Report 3. International Institute for Environment and
Development, London.
Barrett, C. B., Reardon, T., and Webb, P. 2001. Nonfarm Income Diversification and
Household Livelihood Strategies in Rural Africa: Concepts, Dynamics, and Policy
Implications. Food Policy 26(4): 315-331.
Bezabih, M., Köhlin, G., and Mannberg, A. 2011. Trust, Tenure Insecurity and Land
Certification in Rural Ethiopia. Journal of Socio-Economics 40: 833-843
Bezu, S., and Barrett, C., 2012. Employment Dynamics in the Rural Nonfarm Sector in
Ethiopia: Do the Poor Have Time on Their Side? The Journal of Development Studies
48(9): 1223-1240.
Block, S. and Webb, P. 2001. The Dynamics of Livelihood Diversification in Post-famine
Ethiopia. Food Policy 26(4): 333-350.
Crewett, W., Bogale, A., and Korf, B. 2008. Land Tenure in Ethiopia: Continuity and
Change, Shifting Rulers, and the Quest for State Control. International Food Policy
Research Institute.
Davis, J. 2003. The Rural-Non-Farm Economy, Livelihoods and Their Diversification: Issues
and Options. Livelihoods and Their Diversification: Issues and Options.RI report
2783. Natural Resources Institute.The University of Greenwich.
Deininger, K., Jin, S., and Nagarajan, H. K., 2008. Efficiency and Equity Impacts of Rural
Land Rental Restrictions: Evidence from India. European Economic Review 52(5):
892-918.
Deininger, K., Ali, A., and Alemu, T., 2011. Impacts of Land Certification on Tenure
Security, Investment, and Land Market Participation: Evidence from Ethiopia. Land
Economics 87(2): 312-334.
Deininger, K., Jin, S., and Xia. F. 2012. Moving Off the Farm: Land Institutions to Facilitate
Structural Transformation and Agricultural Productivity Growth in China. World
Bank Policy Research Working Paper Series.
Do, Q.-T., and Iyer, L. 2008. Land Titling and Rural Transition in Vietnam. Economic
Development and Cultural Change 56(3) 531-579.
23
Environment for Development
Bezabih et al.
Ellis, F. 2000. The Determinants of Rural Livelihood Diversification in Developing
Countries. Journal of Agricultural Economics 51(2) 289-302.
Holden, S. T., and Yohannes, H. 2002. Land Redistribution, Tenure Insecurity, and Input
Intensity: A Study of Farm Households in Southern Ethiopia. Land Economics 78(4):
573-90.
Holden, S. T., Shiferaw, B. and Pender, J., 2004. Non-farm Income, Household Welfare, and
Sustainable Land Management in a Less-favoured Area in the Ethiopian Highlands.
Food Policy 29(4) 369-392.
Holden, S. T., Deininger, K. D., and Ghebru, H. 2009. Impacts of Low-Cost Land
Certification on Investment and Productivity. American Journal of Agricultural
Economics 91(2): 359-373.
Holden, S. T., Deininger, K. D., and Ghebru, H. 2011. Tenure Insecurity, Gender, Low-cost
Land Certification and Land Rental Market Participation in Ethiopia. The Journal of
Development Studies 47(1) 31-47.
Jacoby, H. G., and Minten, B. 2007. Is Land Titling in Sub-Saharan Africa Cost-Effective?
Evidence from Madagascar. The World Bank Economic Review 21(3) 461-485.
Jalan, J., and Ravallion, R. 2003. Does Piped Water Reduce Diarrhea for Children in Rural
India? Journal of Econometrics 112: 153-73.
Jin, S., and Deininger, K. D. 2009. Land Rental Markets in the Process of Rural Structural
Transformation: Productivity and Equity Impacts from China. Journal of
Comparative Economics 37(4): 629-646.
Kebede, B. 2002. Land Tenure and Common Pool Resources in Rural Ethiopia: A Study
Based on Fifteen Sites. African Development Review 14(1): 113-49.
Lanjouw, J. O., and Lanjouw, P. 2001. The Rural Non-farm Sector: Issues and Evidence from
Developing Countries. Agricultural Economics 26(1) 1-23.
Mundlak, Y. 1978. On the Pooling of Time Series and Cross Section Data. Econometrica 46:
69-85.
Pinckney, T. C., and Kimuyu, P. K. 1994. Land Tenure Reform in East Africa: Good, Bad or
Unimportant? Journal of African Economies 3(1): 1-28.
Rijkers, B., and Söderbom, M. 2013. The Effects of Risk and Shocks on Non-Farm
Enterprise Development in Rural Ethiopia. World Development 35: 119-136.
24
Environment for Development
Bezabih et al.
Rosenbaum, P. R., and Rubin, D. B. 1985. Constructing a Control Group Using Multivariate
Matched Sampling Methods that Incorporate the Propensity Score. Am Stat. 39(1):
33-38.
Shi, X., Heerink, N., and Qu, F. 2007. Choices between Different Off-farm Employment Subcategories: An Empirical Analysis for Jiangxi Province, China. China Economic
Review 18(4): 438-455.
van den Berg, M., and Kumbi, G. E. 2006. Poverty and the Rural Nonfarm Economy in
Oromia, Ethiopia. Agricultural Economics (Supplement) 35: 469-475.
Woldenhana, T., and Oskam, A. 2001. Income Diversification and Entry Barriers: Evidence
from the Tigray Region of Northern Ethiopia. Food Policy 26(4) 351-365.
Wooldridge, J. M. 2002. Econometric Analysis of Cross Section and Panel Data. Cambridge,
Massachusetts: MIT Press.
Zhang, Q. F., Qingguo, M., and Xu, X. 2004. Development of Land Rental Markets in Rural
Zhejiang: Growth of Off-farm Jobs and Institution Building. The China Quarterly
180: 1031-1049.
25
Environment for Development
Bezabih et al.
Figures and Tables
Figure 1. Standardized Bias in Sample Before and After Matching
26
Environment for Development
Bezabih et al.
Figure 2. Bias in Individual Variables Before and After Matching
Figure 3a: Trend in share of individuals in off-farm
activities in certified and non-certified villages
(controlling for covariates): off-farm work (excluding
agricultural off-farm work)
Figure 3b: Trend in share of individuals in off-farm
activities in certified and non-certified villages
(controlling for covariates): All off-farm work
27
Environment for Development
Bezabih et al.
Table 1. Participation in Off-farm Activities
Pooled sample
Certification villages
Non-certification villages
2005
2007
2005
2007
2005
2007
Off-farm
11,79
21,33
15,67
22,10
9,26
20,80
No off-farm activity
88,21
78,67
84,33
77,90
90,74
79,20
Individual activities
Farm work on other land
5,16
1,90
5,83
1,01
4,58
2,56
Free work
91,85
73,80
92,13
67,82
91,60
78,23
Professional work
0,82
4,79
0,58
4,62
1,02
4,91
Unskilled work
2,17
9,94
1,46
9,81
2,80
10,03
Food-for-work
0,14
9,57
0,29
16,74
0,00
4,27
7444
7735
2941
3172
4503
4563
Total (N)
28
Environment for Development
Bezabih et al.
Table 2. Descriptive Statistics of the Variables Used in the Regressions
Pooled sample
Certification villages
Non-certification villages
Mean
Std.dev
Min
Max
Mean
Std.dev
Min
Max
Mean
Std.dev
Min
Max
Age of hh head
50,62
15,01
12,00
105,00
49,40
14,52
16,00
100,00
51,46
15,28
12,00
105,00
Socio-economic variables
Gender of hh head
0,14
0,34
0,00
1,00
0,16
0,37
0,00
1,00
0,12
0,32
0,00
1,00
Hh head is illiterate
0,54
0,50
0,00
1,00
0,51
0,50
0,00
1,00
0,56
0,50
0,00
1,00
Size of household
5,38
2,33
1,00
17,00
5,29
2,32
1,00
14,00
5,44
2,33
1,00
17,00
N. male adults in hh
2,11
1,28
0,00
10,00
1,99
1,22
0,00
6,00
2,19
1,31
0,00
10,00
N female adults in hh
2,02
1,10
0,00
9,00
1,98
1,12
0,00
7,00
2,05
1,09
0,00
9,00
N Livestock (TLU) ny hh
4,60
3,58
0,00
38,72
5,45
4,05
0,00
38,72
4,01
3,08
0,00
21,20
N Oxen by hh
1,80
1,47
0,00
16,00
2,05
1,72
0,00
16,00
1,63
1,24
0,00
7,00
Experience of land loss
0,09
0,28
0,00
1,00
0,06
0,25
0,00
1,00
0,10
0,30
0,00
1,00
Expect loss of land in future
Expect increase in land in
future
0,14
0,35
0,00
1,00
0,13
0,33
0,00
1,00
0,15
0,36
0,00
1,00
0,20
0,40
0,00
1,00
0,19
0,39
0,00
1,00
0,20
0,40
0,00
1,00
Expect no change in land
0,32
0,47
0,00
1,00
0,36
0,48
0,00
1,00
0,29
0,45
0,00
1,00
Uncertain about future land
0,34
0,47
0,00
1,00
0,32
0,47
0,00
1,00
0,36
0,48
0,00
1,00
Land holdings (area in ha)
2,48
1,76
0,01
9,94
2,46
1,83
0,01
9,94
2,49
1,70
0,03
9,94
Mean plot area
0,32
0,22
0,00
2,04
0,31
0,21
0,00
1,72
0,33
0,22
0,00
2,04
Red soil
0,49
0,45
0,00
1,00
0,59
0,44
0,00
1,00
0,43
0,44
0,00
1,00
Black soil
0,42
0,44
0,00
1,00
0,36
0,44
0,00
1,00
0,45
0,44
0,00
1,00
Other soil
0,06
0,23
0,00
1,00
0,02
0,14
0,00
1,00
0,09
0,27
0,00
1,00
Tenure security variables
Plot characteristics
Share of plots with
29
Environment for Development
N
Bezabih et al.
Fertile soil
0,56
0,46
0,00
1,00
0,61
0,46
0,00
1,00
0,53
0,46
0,00
1,00
Medium fertile soil
0,33
0,42
0,00
1,00
0,30
0,42
0,00
1,00
0,35
0,42
0,00
1,00
Non-fertile soil
0,10
0,25
0,00
1,00
0,08
0,22
0,00
1,00
0,11
0,27
0,00
1,00
Flat slope
0,71
0,41
0,00
1,00
0,79
0,37
0,00
1,00
0,65
0,43
0,00
1,00
Steep slope
0,05
0,19
0,00
1,00
0,02
0,13
0,00
1,00
0,06
0,21
0,00
1,00
Medium steep slope
0,24
0,37
0,00
1,00
0,17
0,34
0,00
1,00
0,28
0,39
0,00
1,00
2822
1155
30
1667
Environment for Development
Bezabih et al.
Table 3. PSM logit result: selection into certification kebeles
Hh head is illiterate
Total land area by hh
Mean plot area
N male adults in hh
N livestock owned by hh
N oxen owned by hh
Share of land with black soil
Share of land with red soil
Share of land with flat slope
Share of land with medium slope
Share of land with fertile soil
Share of land with medium fertile soil
Experience of land loss
Experience of land conflict
Pr(Certificate)
-0.053
(0.068)
0.628
(0.501)
-0.829***
(0.301)
-0.115***
(0.032)
0.095***
(0.018)
-0.075**
(0.037)
-0.054
(0.072)
0.262***
(0.068
0.088
(0.084)
-0.320***
(0.092)
-0.154**
(0.064)
-0.158**
(0.064)
-0.401***
(0.115)
0.034
(0.086)
N
1577
LR Chi square (14)
225.71
Pseudo R2
0.106
31
Environment for Development
Bezabih et al.
Table 4. Decision to Participate in Off-farm Activities (Kebele Level Certification). Random Effect Logit
off-farm work (excluding agricultural off-farm work)
All off-farm work
Model I
Hh in treated village
Post treatment
Model II
Model III
Model IV
-2.315***
-2.784***
(0.514)
1.632***
(0.314)
Farm size*treatment
Farm size*post treatment
Model II
Model III
Model IV
-2.789***
-0.903***
-0.869**
-0.931**
(0.635)
(0.639)
(0.333)
(0.407)
(0.411)
3.102***
3.103***
-0.529**
-0.742**
-0.766**
(0.518)
(0.521)
(0.209)
(0.323)
(0.325)
0.374**
0.391**
-0.034
-0.018
(0.159)
(0.161)
(0.104)
(0.105)
-0.781***
-0.798***
0.086
0.082
(0.210)
(0.211)
(0.107)
(0.108)
Experience of land loss
Model I
0.472**
0.461***
(0.231)
(0.176)
Expect decrease in land holdings
0.086
0.060
(Reference is no change)
(0.226)
(0.153)
Expect increase in land holdings
-0.005
-0.075
(0.244)
(0.170)
0.092
-0.138
Uncertain about future land
(0.178)
Year 2007
Total land area by hh
Age of household head
(0.130)
-0.079
-0.640***
-0.652***
-0.609***
1.284***
1.482***
1.486***
1.513***
(0.148)
(0.188)
(0.189)
(0.192)
(0.121)
(0.147)
(0.147)
(0.150)
-0.101
-0.113
-0.087
-0.099
-0.019
-0.016
-0.028
-0.042
(0.097)
(0.101)
(0.104)
(0.105)
(0.064)
(0.064)
(0.068)
(0.069)
-0.025
-0.027
-0.030
-0.027
-0.001
-0.001
-0.001
0.002
(0.034)
(0.035)
(0.035)
(0.036)
(0.027)
(0.027)
(0.027)
(0.027)
32
Environment for Development
Age of head squared
Bezabih et al.
-0.000
-0.000
-0.000
-0.000
-0.000*
-0.000*
-0.000*
-0.000*
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
-0.250
-0.257
-0.315
-0.229
-1.373*
-1.387*
-1.376*
-1.389*
(1.006)
(1.030)
(1.043)
(1.051)
(0.807)
(0.810)
(0.810)
(0.816)
-0.164
-0.170
-0.182
-0.168
0.052
0.046
0.047
0.059
(0.163)
(0.168)
(0.170)
(0.171)
(0.118)
(0.119)
(0.119)
(0.119)
0.292*
0.304**
0.307**
0.322**
0.277**
0.278**
0.280**
0.282**
(0.151)
(0.154)
(0.155)
(0.155)
(0.109)
(0.110)
(0.110)
(0.110)
0.405
0.405
0.421
0.407
0.662*
0.677**
0.676**
0.685**
(0.415)
(0.424)
(0.428)
(0.432)
(0.342)
(0.343)
(0.343)
(0.346)
-0.013
-0.013
-0.013
-0.014
-0.014*
-0.014*
-0.014*
-0.014*
(0.011)
(0.012)
(0.012)
(0.012)
(0.008)
(0.008)
(0.008)
(0.008)
-0.038
-0.038
-0.039
-0.038
-0.073**
-0.075**
-0.075**
-0.076**
(0.042)
(0.043)
(0.043)
(0.044)
(0.035)
(0.035)
(0.035)
(0.035)
-0.157
-0.125
-0.126
-0.127
0.040
0.028
0.026
0.031
(0.133)
(0.137)
(0.139)
(0.139)
(0.091)
(0.092)
(0.092)
(0.092)
0.023
0.018
0.023
0.023
0.025
0.029
0.028
0.028
(0.057)
(0.059)
(0.060)
(0.060)
(0.039)
(0.039)
(0.039)
(0.039)
Share of plots: red soil
0.675**
0.619*
0.631*
0.657*
-0.068
-0.051
-0.057
-0.017
(Other color is reference)
(0.342)
(0.354)
(0.356)
(0.359)
(0.250)
(0.252)
(0.252)
(0.254)
share of plots: black soil
0.157
0.164
0.145
0.145
-0.091
-0.090
-0.091
-0.069
(0.336)
(0.347)
(0.350)
(0.352)
(0.248)
(0.250)
(0.250)
(0.251)
Share of plots: non fertile soil
-0.152
-0.157
-0.179
-0.195
0.242
0.245
0.248
0.213
(Fertile soil is reference)
(0.433)
(0.446)
(0.449)
(0.451)
(0.290)
(0.292)
(0.292)
(0.294)
Share of plots: medium fertile soil
-0.127
-0.060
-0.057
-0.102
0.035
0.014
0.015
-0.019
(0.221)
(0.228)
(0.231)
(0.233)
(0.166)
(0.167)
(0.167)
(0.168)
Share of plots: steep slope
-0.761
-0.679
-0.623
-0.622
-0.268
-0.265
-0.269
-0.258
(Flat slope is reference)
(0.521)
(0.537)
(0.543)
(0.545)
(0.368)
(0.371)
(0.370)
(0.371)
Share of plots: medium steep slope
0.062
-0.007
-0.002
-0.003
-0.095
-0.071
-0.068
-0.086
(0.257)
(0.265)
(0.268)
(0.269)
(0.187)
(0.188)
(0.188)
(0.189)
Hh head is female
Hh head is illiterate
N male adults in hh
N female adults in hh
N male adults^2
N female adults^2
N oxen owned by hh
N livestock owned by hh
33
Environment for Development
Mean plot area
Bezabih et al.
0.061
-0.028
-0.020
0.068
-0.428
-0.380
-0.362
-0.252
(0.662)
(0.692)
(0.697)
(0.702)
(0.451)
(0.453)
(0.453)
(0.456)
-1.503
0.243
0.202
-0.031
-0.966
0.081
0.098
0.016
(1.031)
(1.032)
(1.045)
(1.063)
(0.748)
(0.754)
(0.758)
(0.774)
-1.251
-0.806
-0.681
-0.662
-0.833
-0.784
-0.785
-0.755
(1.091)
(0.786)
(0.736)
(0.729)
(0.517)
(0.499)
(0.499)
(0.492)
Kebele fixed effects
YES
YES
YES
YES
YES
YES
YES
YES
Chamberlain-Mundlak effects
YES
YES
YES
YES
YES
YES
YES
YES
Chi-squared
138.832
144.847
146.312
147.440
293.059
292.891
293.340
293.103
N
2478
2478
2478
2478
2478
2478
2478
2478
Constant
lnsig2u constant
34
Environment for Development
Bezabih et al.
Table 5. Multinominal Logit Results: Activity Choice (Kebele Level Certification). Predictive Margins
Hh in treated village
Year 2007
Post treatment
Total land area by hh
Farm size*post treatment
Age of household head
Hh head is female
Hh head is illiterate
N male adults in hh
N female adults in hh
N oxen owned by hh
N livestock owned by hh
35
Farm worker
Free worker
Professional
Unskilled
FFW
-0.001
0.014
0.018
-0.031
0.004
(0.006)
(0.028)
(0.017)
(0.030)
(0.004)
-0.003
0.106***
0.042**
0.041***
0.050***
(0.006)
(0.022)
(0.020)
(0.008)
(0.007)
-0.001
0.030
-0.035*
0.071*
0.031***
(0.012)
(0.048)
(0.019)
(0.031)
(0.010)
-0.001
0.013
-0.004
-0.003
-0.012**
(0.004)
(0.010)
(0.005)
(0.005)
(0.005)
-0.002
-0.009
0.005
-0.014*
-0.004
(0.005)
(0.012)
(0.004)
(0.008)
(0.005)
-0.000**
-0.005***
-0.000
-0.000
-0.000
(0.000)
(0.001)
(0.000)
(0.000)
(0.000)
-0.004
-0.112
0.031*
0.019
0.003
(0.011)
(0.069)
(0.017)
(0.027)
(0.016)
0.008
0.015
-0.009*
-0.002
-0.011
(0.005)
(0.019)
(0.005)
(0.007)
(0.007)
0.000
-0.000
0.001
0.005**
0.005***
(0.002)
(0.005)
(0.001)
(0.002)
(0.002)
0.001
-0.011
-0.016***
-0.003
0.008***
(0.002)
(0.014)
(0.003)
(0.006)
(0.003)
-0.005*
0.058***
-0.003
-0.009*
-0.012**
(0.003)
(0.010)
(0.003)
(0.005)
(0.005)
0.001
0.001
0.000
-0.001
0.001
(0.001)
(0.004)
(0.001)
(0.002)
(0.002)
Environment for Development
Bezabih et al.
Share of plots: red soil
0.010
-0.005
-0.012**
0.001
-0.002
(Other color is reference)
(0.015)
(0.032)
(0.006)
(0.012)
(0.011)
share of plots: black soil
0.020
-0.023
-0.012**
0.002
0.008
(0.015)
(0.032)
(0.006)
(0.012)
(0.011)
-0.014
-0.045
0.006
0.031
0.021
(0.025)
(0.070)
(0.031)
(0.027)
(0.030)
Share of plots: non fertile soil
-0.012
-0.023
-0.008
-0.025
0.001
(Fertile soil is reference)
(0.012)
(0.036)
(0.011)
(0.015)
(0.015)
Share of plots: medium fertile soil
-0.002
-0.018
-0.003
-0.010
0.004
(0.006)
(0.022)
(0.005)
(0.008)
(0.007)
Share of plots: steep slope
0.011
-0.009
0.001
-0.017
0.005
(Flat slope is reference)
(0.010)
(0.044)
(0.009)
(0.022)
(0.018)
Share of plots: medium steep slope
-0.016
0.001
-0.005
0.003
0.010
(0.010)
(0.023)
(0.006)
(0.008)
(0.007)
Mean plot area
Wald chi2(100)
20469.14
Pseudo R2
0.155
N
2610
36
Environment for Development
Bezabih et al.
Table 6. Trend Analysis: Placebo Effects
off-farm work (excluding agricultural off-farm
work)
All off-farm work
KEBELE LEVEL
CERT
I
Placebo 2000
0.332
-0.821***
(0.472)
(0.225)
II
Placebo 2002
III
IV
I
0.785***
III
IV
-0.142
(0.299)
Placebo 2005
II
(0.216)
-1.209***
-0.826***
0.819***
0.775***
(0.340)
(0.294)
(0.206)
(0.202)
Post treatment
2.203***
0.379
(0.424)
(0.285)
Chi-squared
146.904
150.920
153.018
222.540
364.708
359.033
366.071
560.804
N
3193
3193
3193
4350
3193
3193
3193
4350
37
Environment for Development
Bezabih et al.
Appendix Figures
Figure A1. Standardized Bias in Sample Before and After Matching. Identification
Based on Household Certification.
38
Environment for Development
Bezabih et al.
Figure A2. Bias in Individual Variables Before and After Matching. Identification Based
on Household Certification.
39
Environment for Development
Bezabih et al.
Appendix Tables
Table A1. Household Level Certification: Participation Results
off-farm work (excluding agricultural off-farm work)
All off-farm work
Model I
Certified hh
Post-treatment
Model II
Model III
Model IV
-0.944**
-1.460**
(0.403)
(0.598)
0.373
(0.444)
Farm size*certified hh
Farm size*post treatment
Model II
Model III
Model IV
-1.429**
0.221
0.325
0.320
(0.603)
(0.294)
(0.438)
(0.441)
1.085*
1.126*
-0.530*
-0.824*
-0.804*
(0.599)
(0.604)
(0.310)
(0.428)
(0.433)
0.269
0.242
-0.055
-0.053
(0.201)
(0.204)
(0.143)
(0.144)
-0.376*
-0.352*
0.124
0.123
(0.207)
(0.209)
(0.132)
(0.133)
Experience of land loss
Model I
0.567
0.132
(0.350)
(0.257)
Expect decrease in land holdings
0.184
0.028
(Reference is no change)
(0.342)
(0.221)
Expect increase in land holdings
0.221
-0.129
(0.376)
(0.255)
0.152
-0.164
(0.287)
(0.201)
Uncertain about future land holdings
Year 2007
Total land area by hh
-0.226
-0.452
-0.449
-0.405
1.368***
1.685***
1.696***
1.666***
(0.235)
(0.346)
(0.346)
(0.352)
(0.182)
(0.267)
(0.268)
(0.271)
-0.060
-0.053
-0.035
-0.061
0.016
0.021
-0.012
-0.021
40
Environment for Development
Bezabih et al.
(0.167)
(0.165)
(0.185)
(0.187)
(0.093)
(0.094)
(0.110)
(0.111)
-0.061
-0.050
-0.047
-0.047
0.006
0.006
0.005
0.004
(0.055)
(0.055)
(0.055)
(0.055)
(0.041)
(0.041)
(0.041)
(0.041)
0.000
0.000
0.000
0.000
-0.001
-0.001
-0.001
-0.000
(0.001)
(0.001)
(0.001)
(0.001)
(0.000)
(0.000)
(0.000)
(0.000)
-0.213
0.346
0.429
0.399
-0.788
-0.746
-0.721
-0.747
(1.939)
(1.937)
(1.928)
(1.940)
(1.385)
(1.392)
(1.391)
(1.401)
-0.090
-0.014
-0.012
0.005
0.243
0.246
0.246
0.240
(0.260)
(0.263)
(0.261)
(0.262)
(0.171)
(0.172)
(0.172)
(0.173)
0.426
0.501*
0.485
0.519*
0.309*
0.317*
0.336*
0.347*
(0.296)
(0.298)
(0.296)
(0.296)
(0.177)
(0.180)
(0.181)
(0.182)
0.559
0.361
0.294
0.343
0.323
0.316
0.336
0.354
(0.778)
(0.777)
(0.774)
(0.779)
(0.569)
(0.572)
(0.571)
(0.576)
-0.020
-0.027
-0.027
-0.029
-0.018
-0.019
-0.020
-0.021
(0.023)
(0.023)
(0.023)
(0.023)
(0.013)
(0.014)
(0.014)
(0.014)
-0.034
-0.016
-0.010
-0.014
-0.030
-0.029
-0.031
-0.033
(0.074)
(0.074)
(0.074)
(0.074)
(0.054)
(0.055)
(0.055)
(0.055)
-0.202
-0.226
-0.242
-0.235
0.171
0.168
0.176
0.178
(0.222)
(0.223)
(0.223)
(0.224)
(0.138)
(0.138)
(0.139)
(0.140)
0.010
0.021
0.017
0.015
-0.019
-0.017
-0.018
-0.018
(0.097)
(0.097)
(0.098)
(0.098)
(0.061)
(0.061)
(0.061)
(0.061)
Share of plots: red soil
1.055*
1.092*
1.111*
1.179**
0.125
0.154
0.133
0.163
(Other color is reference)
(0.594)
(0.596)
(0.594)
(0.598)
(0.383)
(0.385)
(0.386)
(0.387)
share of plots: black soil
1.187**
1.188**
1.235**
1.250**
0.425
0.450
0.439
0.462
(0.559)
(0.559)
(0.558)
(0.561)
(0.372)
(0.375)
(0.375)
(0.376)
Share of plots: non fertile soil
0.092
0.071
-0.110
0.053
0.157
0.177
0.217
0.272
(Fertile soil is reference)
(1.189)
(1.175)
(1.202)
(1.213)
(0.683)
(0.686)
(0.693)
(0.698)
Share of plots: medium fertile soil
-0.289
-0.379
-0.374
-0.380
0.359
0.418
0.418
0.399
(0.661)
(0.666)
(0.665)
(0.667)
(0.405)
(0.409)
(0.410)
(0.412)
Age of household head
Age of head squared
Hh head is female
Hh head is illiterate
N male adults in hh
N female adults in hh
N male adults^2
N female adults^2
N oxen owned by hh
N livestock owned by hh
41
Environment for Development
Bezabih et al.
Share of plots: steep slope
-0.688*
-0.752**
-0.741**
-0.761**
-0.127
-0.137
-0.131
-0.142
(Flat slope is reference)
(0.362)
(0.364)
(0.362)
(0.364)
(0.250)
(0.250)
(0.251)
(0.253)
Share of plots: medium steep slope
-0.405
-0.231
-0.163
-0.277
-0.267
-0.251
-0.260
-0.250
(0.770)
(0.769)
(0.765)
(0.773)
(0.503)
(0.505)
(0.506)
(0.509)
0.389
0.426
0.379
0.323
0.349
0.344
0.364
0.341
(0.423)
(0.424)
(0.425)
(0.426)
(0.275)
(0.275)
(0.276)
(0.279)
-1.382
-0.929
-0.742
-1.160
-0.441
-0.562
-0.600
-0.574
(1.789)
(1.795)
(1.787)
(1.807)
(1.178)
(1.195)
(1.210)
(1.231)
-1.261
-1.409
-1.705
-1.882
-1.641
-1.593
-1.568
-1.518
(1.844)
(2.038)
(2.654)
(3.149)
(1.533)
(1.476)
(1.446)
(1.385)
Kebele fixed effects
YES
YES
YES
YES
YES
YES
YES
YES
Chamberlain-Mundlak effects
YES
YES
YES
YES
YES
YES
YES
YES
Chi-squared
61.236
64.562
67.694
69.717
145.446
143.500
143.471
143.332
Mean plot area
Constant
lnsig2u constant
N
1115
42
1115
1115
1115
1115
1115
1115
1115
Environment for Development
Bezabih et al.
Table A2. Household Level Certification: Activity Choice
Certificate hh
Year 2007
Post treatment
Total land area by hh
Farm size*post-treatment
Age of household head
Hh head is female
Hh head is illiterate
N male adults in hh
N female adults in hh
N oxen owned by hh
N livestock owned by hh
Share of plots: non fertile soil
Share of plots: steep slope
Free work
Unskilled
FFW
0.303***
-0.418***
0.034***
(0.045)
(0.069)
(0.011)
0.149***
0.035***
0.046***
(0.035)
(0.009)
(0.010)
-0.150**
0.439***
-0.002
(0.069)
(0.075)
(0.024)
0.011
-0.001
-0.015
(0.012)
(0.007)
(0.013)
-0.004
-0.004
-0.000
(0.016)
(0.008)
(0.013)
-0.005***
-0.000
0.000
(0.001)
(0.000)
(0.000)
-0.053
-0.003
-0.025
(0.104)
(0.033)
(0.026)
0.038
-0.001
-0.009
(0.029)
(0.012)
(0.011)
-0.000
0.006**
0.003
(0.007)
(0.003)
(0.003)
-0.026
0.006
0.010**
(0.024)
(0.006)
(0.004)
0.071***
-0.013*
-0.015*
(0.016)
(0.007)
(0.008)
-0.005
0.001
0.003
(0.006)
(0.004)
(0.003)
0.020
-0.019
0.007
(0.048)
(0.021)
(0.019)
0.037
-0.014
-0.004
(0.056)
(0.032)
(0.029)
Wald chi2(100)
11778.63
Pseudo R2
0,1663
N
1149
43
Environment for Development
Bezabih
Table A3. Household Level Certification: Trend Analysis
off-farm work
(excluding agricultural
off-farm work)
All off-farm work
HH LEVEL CERT
I
Placebo 2000
-1.036***
-0.743***
(0.390)
(0.284)
Placebo 2002
II
III
IV
I
II
0.649*
-0.415
(0.394)
(0.287)
Placebo 2005
III
IV
0.543
0.482
1.252***
1.222***
(0.386)
(0.373)
(0.290)
(0.287)
Post treatment
1.354***
0.777**
(0.493)
(0.330)
Chi-squared
78.725
75.553
76.712
103.127
211.382
208.651
213.667
280.098
N
1356
1356
1356
1933
1356
1356
1356
1933
44